Unsupervised Physics Informed Decomposition of Incomplete Time-Resolved Spectroscopy

Abstract

Raman spectroscopy is often hindered by strong autofluorescence backgrounds and noise. While methods for computational removal exist, better results are often achieved through a time-intensive process of waiting for fluorescence to bleach before recording the Raman spectrum. Here, we instead show that short, incomplete recordings of the bleaching process already contain enough information to recover the underlying Raman signal. We present the first unsupervised model that operates on short, incomplete, variable-length sequences of time-resolved measurements, decomposing them into Raman spectra and autofluorescence components while simultaneously removing measurement noise. Our VAE-based method simultaneously learns (i) a bank of fluorophore spectra comprising the baseline, (ii) possible peak positions and shapes, (iii) a model of the measurement noise, and (iv) neural networks for the decomposition of Raman spectra and decay behavior. We evaluate our method on spectral reconstruction and downstream peak detection, where it outperforms state-of-the-art approaches on multiple simulated benchmarks as well as on a newly collected real-world dataset of spectral time series. We further show that normally disregarded fluorescence decay dynamics themselves may contain discriminative information that may be exploited in future work.